Computer Vision Engineer, 3D

Mecka AI is building the data infrastructure layer for robotics and embodied AI, designing and operating global systems for data capture, data labeling, and hardware-enabled workflows used to train and validate humanoid and embodied AI systems. The role is a coding-heavy computer vision engineer position owning the classical side of the stack: traditional computer vision, temporal filtering, and inverse kinematics that turn raw per-frame model predictions into clean, rigid, smooth, correctly-calibrated spatial data. The immediate focus is the 3D hands pipeline, with the engineer later acting as the classical-CV specialist across the organization.

Responsibilities

Own the classical 3D geometry that places reconstructions correctly in space: multi-view geometry, triangulation, PnP, camera calibration, undistortion, and reprojection; Design and tune temporal filters (low-pass, Kalman-family, interpolation) that remove jitter without losing real motion; Run inverse kinematics and enforce physical constraints such as bone-length rigidity so reconstructed skeletons stay valid; Build the checks that keep bad frames out of delivery; Turn model output into clean delivered data, including per-frame results, reports, and the QA metrics that gate delivery, and keep the pipeline running at scale; Build visualizers and dashboards to inspect trajectories, overlays, and failure cases; Be the go-to engineer for classical computer-vision problems across the pipeline as they come up; Partner closely with research teams without training the models; Focus initially on the 3D hands pipeline

Requirements

5+ years of professional experience in software engineering and computer vision; Strong software engineering with clean, efficient, production Python (C++ is a plus); Solid multi-view geometry: triangulation, PnP, camera calibration, reprojection, and coordinate transforms; Temporal filtering and signal processing: low-pass and Kalman-family filters, interpolation, and a feel for the trade-off between smoothness and fidelity; Inverse kinematics and rigid-body or bone-length constraints on articulated structures; Enough ML to work with deep models such as pose and mesh regressors, including their outputs and failure modes, without training them yourself; A data-quality mindset: comfortable with noisy real-world data and building QA that catches problems before customers do

Nice to have

Human pose or mesh reconstruction (hand or body), including keypoint and mesh pipelines; Depth in Bayesian filtering or state estimation; experience with optimization frameworks such as Ceres or GTSAM; Building CV tooling such as trajectory visualizers and overlay or QA dashboards; Stereo, multi-view, or ego-exo capture experience; Light temporal ML, such as small networks for smoothing or gap-filling